Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20220225155552.30636-1-graf@amazon.com> <CAMj1kXGtANm3SMoREymDSyx+wpn3L=Ex5q5mpgQigOwmEp33Lg@mail.gmail.com>
 <YhkQKfE8ErtFBmSB@zx2c4.com>
In-Reply-To: <YhkQKfE8ErtFBmSB@zx2c4.com>
From:   Ard Biesheuvel <ardb@kernel.org>
Date:   Fri, 25 Feb 2022 18:30:14 +0100
Message-ID: <CAMj1kXEtUUod8Hp6VhS6k7iDKYkFj_t_J=qS2XF1p2X_SFdTvg@mail.gmail.com>
Subject: Re: [PATCH] ACPI: bus: Match first 9 bytes of device IDs
To:     "Jason A. Donenfeld" <Jason@zx2c4.com>
Cc:     Alexander Graf <graf@amazon.com>,
        ACPI Devel Maling List <linux-acpi@vger.kernel.org>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        Len Brown <lenb@kernel.org>,
        "Rafael J. Wysocki" <rafael@kernel.org>,
        Greg KH <gregkh@linuxfoundation.org>,
        "Woodhouse, David" <dwmw@amazon.co.uk>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Fri, 25 Feb 2022 at 18:21, Jason A. Donenfeld <Jason@zx2c4.com> wrote:
>
> On Fri, Feb 25, 2022 at 6:13 PM Ard Biesheuvel <ardb@kernel.org> wrote:
> >
> > The device subsystem side of things already deals with this properly:
> > the modalias of the QEMU vmgenid device comes up as
> > 'acpi:QEMUVGID:VM_GEN_COUNTER', which means it already captures the
> > entire string, and exposes it in the correct way (modulo the all caps)
>
> Ahh, so the userspace side of this won't work right. Shucks. That's what
> I was concerned about.
>
> > I don't like this hack. If we are going to accept the fact that CIDs
> > could be arbitrary length strings, we should handle them properly.
> >
> > So what we need is a way for a module to describe its compatibility
> > with such a _CID, which shouldn't be that complicated.
>
> Can't we do something more boring and just...
>
> diff --git a/include/linux/mod_devicetable.h b/include/linux/mod_devicetable.h
> index 4bb71979a8fd..5da5d990ff58 100644
> --- a/include/linux/mod_devicetable.h
> +++ b/include/linux/mod_devicetable.h
> @@ -210,9 +210,9 @@ struct css_device_id {
>         __u8 type; /* subchannel type */
>         kernel_ulong_t driver_data;
>  };
>
> -#define ACPI_ID_LEN    9
> +#define ACPI_ID_LEN    16
>
>  struct acpi_device_id {
>         __u8 id[ACPI_ID_LEN];
>         kernel_ulong_t driver_data;
>
>
> As you can see from the context, those additional 7 bytes were being
> wasted on padding anyway inside the acpi_device_id struct, so it's
> basically free, it would seem. This seems like the least convoluted way
> of solving this issue? If we ever encounter _more_ ACPI devices with
> weird names, we could revisit a fancy dynamic solution, but for now, why
> don't we keep it simple?
>

Yeah, good point. I think this is fine, although there are a few other
uses of ACPI_ID_LEN in the tree. So perhaps this should be something
like

#define ACPI_ID_LEN    9
#define ACPI_CID_LEN    16

/* explanation goes here */

struct acpi_device_id {
    __u8 id[ACPI_CID_LEN];

instead? At a quick glance, none of those ACPI_ID_LEN users seem
related to the CID or the match metadata.